Fig. 6

Expected specificity of eCLIP datasets with regard to various features. a The left heatmap displays all eCLIP datasets, clustered based on their similarity scores and recall into 7 clusters. For each dataset, the heatmap also shows its enrichment in the mRNA interactome proteomics (log2FC eRIC), the number of KH or RRM in RBP, the average PEKA score across the top 50 ranked k-mers, and the number of tXn in “protein-coding” region (n tXn). Values shown on this heatmap are available in Additional file 7: Table S6. Grayscale heatmaps from left to right show the mean nucleotide content across the top 50 ranked 5-mers for each dataset in PEKA (left) and in vitro data (middle) and % of thresholded crosslinks derived from each transcript region (right). b Violin plots show quantitative distribution of features within clusters. In addition to features presented in the heatmap (recall, similarity score, log2FC eRIC, mean PEKA score across top 50 k-mers, number of tXn, and number of KH or RRM domains), they also show the number of intrinsically disordered domains (IDRs) and the total number of compositional biases for RBPs within each cluster. c, d Stacked bar plots showing the frequency of a particular domain (c) and compositional biases (d) per RBP within each cluster. Frequency is expressed as a total count of domain or compositional bias per RBP. Detailed structural information for the analyzed RBPs is available in Additional file 10: Table S9